Superwhisper has launched the S1 household of fashions: S1-Voice, S1-Language, and S1-mini. S1-Voice is a cloud speech-to-text mannequin, and S1-Language is a cloud instruction-following mannequin for cleanup and formatting. The one that’s fairly attention-grabbing outdoors the app is S1-mini, launched with open weights on Hugging Face. S1-mini is a 0.6B textual content normalizer, not a transcriber and never a chat mannequin. It sits after automated speech recognition and rewrites uncooked transcripts as clear written textual content: fillers eliminated, self-corrections resolved to what the speaker landed on, punctuation and capitalization utilized, and spoken numbers, dates, foreign money, and e-mail addresses rendered in written type. It’s fine-tuned from Qwen/Qwen3-0.6B, covers English solely in launch v1, and is steered solely by a three-axis management line positioned above the transcript. Superwhisper stories 94.8% token accuracy on a held-out set of seven,519 circumstances, measured grasping on the quantized construct.
Is it deployable?
Sure, however solely S1-mini. S1-mini is revealed on Hugging Face beneath Apache 2.0 plus a naming clause. S1-Voice and S1-Language are Superwhisper-hosted providers, so they’re consumable, not self-hostable.
- Firm stage: Any stage. The Q4_K_M GGUF construct is a 462 MB file that runs on a laptop computer CPU. Solo builders can ship it inside a desktop app. Enterprises can run it behind a VPC the place audio transcripts can not go away the community.
- Industries: Healthcare and scientific documentation, authorized, monetary providers, buyer help, developer tooling, accessibility and dwell captioning.
- Functions: Dictation apps, meeting-notes instruments, dwell captioning, voice-driven editors, voice-to-CRM entry, and any pipeline that turns uncooked ASR output into textual content a human will learn.
What S1-mini does
S1-mini is a textual content normalizer, not a transcriber and never a chat mannequin. It sits after automated speech recognition:
audio → ASR (Whisper, Parakeet, …) → S1-mini → clear textual content
It removes filler phrases, resolves false begins and self-corrections to the worth the speaker landed on, applies punctuation and capitalization, and renders spoken numbers, dates, instances, foreign money, and e-mail addresses in written type. Say “help at superwhisper dot com” and also you get [email protected].
The mannequin is fine-tuned from Qwen/Qwen3-0.6B. It has 596M distinctive parameters (0.44B non-embedding), 28 layers, 16 question heads and eight key/worth heads with GQA, and BF16 weights. The Hub sidebar stories 0.8B as a result of the tied embedding is saved twice; the cardboard explains the discrepancy explicitly. Launch v1 covers English solely, and advisable enter is roughly 1,000 tokens.
The management line is your entire interface
S1-mini takes a set system immediate, then a management line, then the uncooked transcript:
[Styling: ] [Structure: ] [Context: ]
Styling takes informal, semi-casual, semi-formal, or formal. Construction takes prose or lists. Context takes basic or e-mail. All three axes are unbiased, and each mixture was skilled. Ship values outdoors these units, or reword the system immediate, and output can degrade or garble. Observe the small mismatch value understanding: the Superwhisper app exposes a five-stop tone slider that provides a “balanced” preset, whereas the open weights doc 4 skilled Styling values.
The mannequin can also be constrained by design. It doesn’t add content material you didn’t say, appropriate info, soften profanity, or rewrite dialect. Filler-only enter returns an empty string, and integrations ought to deal with that as a sound outcome.
Two settings that break most integrations
First, enable_thinking=False is required. The chat template is Qwen3’s, unchanged, and Qwen3 defaults to pondering on. S1-mini was skilled with pondering off, so the assistant flip should open with an empty
Second, decode greedily. generation_config.json ships do_sample: false. The GGUF builds nonetheless carry Qwen3’s inherited temp = 0.6, top_p = 0.95, and top_k = 20 metadata, so go temperature 0 explicitly on each request. In llama.cpp, use --jinja with --chat-template-kwargs '{"enable_thinking":false}' somewhat than --reasoning-budget 0, which degrades output.
Reported analysis
Superwhisper evaluated S1-mini on a held-out set of seven,519 circumstances throughout 104 transcripts. Token accuracy is 94.8%, measured grasping on the Q4_K_M construct, with a text-edit error price of 11.6%. On email-formatted textual content it identifies the greeting line 99.3% of the time and the sign-off 97.9%. It matches the right output construction, checklist versus paragraph, 97.6% of the time, and produces actual e-mail addresses in 92% of circumstances. Fewer than 1% of generations present looping or truncation, and the mannequin appropriately withholds output 98.6% of the time when nothing needs to be transcribed. These are vendor-reported numbers on an inner take a look at set, not third-party outcomes.
The 2 cloud fashions
S1-Voice is the hosted speech-to-text mannequin. Superwhisper stories transcription as much as 46x quicker than talking time, with most dictations beneath 30 seconds showing 0.32 seconds after you cease. Throughout eight datasets together with assembly audio and earnings calls, it averages 6.8% phrase error price and drops to 2.2% on LibriSpeech. Superwhisper says that 6.8% common was the bottom of 15 fashions it examined, and that S1-Voice scored 83 out of 100 on its blended metric towards WisprFlow’s 76.
S1-Language is the hosted instruction-following mannequin for cleanup, formatting, and summarization, and it seems within the mannequin picker alongside fashions from Anthropic, OpenAI, and Groq. The advisable defaults are Cohere Transcribe plus S1-mini offline, or S1-Voice plus S1-Language within the cloud.
Interactive explainer
The embed under permits you to change every control-line axis and watch the output change. Each enter/output pair is taken verbatim from the mannequin card.
Key Takeaways
- S1-mini is a 0.6B open-weights textual content normalizer for ASR output, not a transcriber or chat mannequin.
- The Q4_K_M GGUF is 462 MB and runs on a laptop computer CPU, so on-device deployment is lifelike.
- A set system immediate plus a three-axis management line is the one steering mechanism.
enable_thinking=Falseand temperature 0 are obligatory; most integration bugs hint to those.
Try the Mannequin Weights and Technical Particulars. Additionally, be happy to observe us on Twitter and don’t overlook to affix our 150k+ML SubReddit and Subscribe to our Publication. Wait! are you on telegram? now you may be a part of us on telegram as properly.
Must companion with us for selling your GitHub Repo OR Hugging Face Web page OR Product Launch OR Webinar and so on.? Join with us
Michal Sutter is an information science skilled with a Grasp of Science in Information Science from the College of Padova. With a stable basis in statistical evaluation, machine studying, and knowledge engineering, Michal excels at reworking advanced datasets into actionable insights.
